Claude Fable 5 will be available again globally tomorrow.
After a series of productive conversations with the US government, we're redeploying the model with a new set of classifiers to target and block more cybersecurity tasks. In the near term, some routine tasks like coding and debugging will fall back to Opus 4.8. We’ll continue to refine these classifiers over the coming weeks to reduce false positives and better distinguish genuine misuse from legitimate requests.
We’ve also begun drafting a consensus framework—with Amazon, Microsoft, Google, and other Glasswing partners—for assessing the severity of AI jailbreaks and how AI developers should respond to them. We invite other industry partners and model providers to join us in this effort.
Finally, we’re scaling up our collaboration with the US government on model testing and safeguards. This will include pre-release access to models and safeguards for evaluation, information sharing on jailbreaks and misuse, and dedicated resources for joint research.
Thank you to our users for your patience, and to our partners across the government, industry, and the research community who worked alongside us to make Fable 5 available again.
Read our full blog: https://t.co/VHyum831ri
🚨 BREAKING: U.S. government will decide who gets access to GPT-5.6
OpenAI will release GPT-5.6 only in a limited preview to a small group of partners.
Sam Altman told staff the government would be "approving access customer by customer."
Commerce Sec Lutnick personally called Altman warning: don't launch without approvals from other agencies.
A de facto licensing regime.
ITS HAPPENING
building inside claude an app using deepseek AI is really difficult, because claude tries everything to make deepseek look bad. But today, there was finally an eval I ran where DeepSeek was clearly winning :D
it was about having stronger guardrails to do websearches to ground knowledge even though its in the training set already
in this case about luggage policy of Japan Transocean Air
whats your experience with claude building apps using other models?
the all-cloudflare stack unlocked something: i can run ~10 dev sessions concurrently on one machine.
no docker. D1 + R2 + everything local via wrangler. so the only ceiling is WSL/RAM, not "how many postgres containers fit in memory."
the one friction left: ports. i tell claude to grab a random free port (5300–5400) per worktree and check it's free before starting. works, but feels hacky — and there's a race in "check then bind."
how are you handling per-worktree ports for concurrent local dev? deterministic hash → port? 🤔
Claude design is great to start. But I have my problems with extending designs in Claude Design so far. First shot great.
It is not very clear about blocks, components, pages, templates and etc.
That leads to divergence already in there. So, I dropped it
Anyone can point me how to use Claude Design "properly"? So, how to prompt to get a proper design system out
@mreflow interesting that they all go to hardware. must be tough to enter. seems unavoidable to become truly infrastructure. and ultimatly not obsolete
software is not enough anymore it seems
Anyone else expericing the need to "revise a plan" to enforce YOLO mode all the way to the end.
I always click on revise, and kindly state to please go through 6 PRs in one go and tell me afterwards assumptions and decisions which were done.
Saved me a lot of round trips - and that despite already YOLOing everything
Wow, that is so well deserved. Intercom grew with the market dynamics, demand and possibilities of pushing technical boundaries.
I saw presentations at the Lisbon web summit from one of your co-founders and it was 3 years ago already talking about agentic customer support/service age.
👏👏
the time we arrive in is just -- crazy?
right now, in parallel:
— 5 coding tasks running
— a Cowork task digesting last month's bank transactions into my budget sheet
— a Claude + Chrome MCP session filling my 2025 numbers into the Danish tax portal
sitting alone in my room laughing. it's so good it's ridiculous. 😱
i used to pay a tax advisor ~a year of Claude Max for this. and i still had to gather everything myself across 4 banks ANYWAYS! 😅
what's the last AI workflow that made you laugh because it shouldn't be possible yet?
@KrishnaBuildsAI But it's true that given the camera support in phone today it gets very close. The delta gets so small that the economic margin for a photographer is getting thin
Nice. I believe in a two brain part system. Doing both manual notes and then having a second brain digest constantly coloated to your own brain - your obsidian vault. That why I made https://t.co/WmIcgHkE5l
I think we need predigestion and not always run with fresh context gathering
Wow, Jensen is for me inspirational. Really adopting as fast as possible anything what is there to expand humanity's intelligence even faster
I admire him!
We have been working closely with @nvidia to ensure Hermes Agent works smoothly on their new @NVIDIARTXSpark superchip and integrates with the new OpenShell runtime, which connects Hermes to @Microsoft's security primitives.
Watch our feature in the big announcement at Computex:
@NousResearch@nvidia@NVIDIARTXSpark@Microsoft Jensen is inspirational. Pushing with Nvidia every day and accepting all these efforts on all front. Seems like one of the most agnostic people in the space.
I admire him